Twenty-one days to a habit

The claim is that twenty-one days of repetition establishes a habit. It appears in training programmes, productivity writing and health promotion material, usually stated as an established finding.

1. Present day

   The claim in circulation

   Stated as a fixed interval, frequently with the implication that research established it.

2. 1960 onward

   Repetition through the self-improvement literature

   Carried forward across six decades of books, courses and articles, each generally citing the previous rather than the origin.

3. Terminus

   Maxwell Maltz, Psycho-Cybernetics

   A cosmetic surgeon writing about self-image. His observation was that patients appeared to require a minimum of about twenty-one days to adjust to an altered appearance, or to the absence of a limb. It was clinical impression, not measurement, and its subject was adjustment to a changed body rather than the acquisition of a behaviour.

The qualifier that was dropped

Maltz's wording described a minimum of approximately twenty-one days. Two hedges sit in that phrase, and both were removed as the claim travelled: a minimum became a duration, and an approximation became a fixed figure. Neither the book nor its author claimed otherwise; the book was not presenting an experimental result and did not pretend to.

What the research actually shows

The most frequently cited empirical work on the question was published in 2010 by Phillippa Lally, Cornelia van Jaarsveld, Henry Potts and Jane Wardle in the European Journal of Social Psychology. Participants selected an eating, drinking or activity behaviour to perform daily in a consistent context, and rated its automaticity over twelve weeks. The researchers modelled the time taken to reach a plateau.

The median was around sixty-six days, and the range across individuals and behaviours was very wide — from under three weeks at one end to well over half a year at the other. The variation, not the median, is the finding. Simpler behaviours reached automaticity faster than more demanding ones.

Limitations of the replacement figure

The sixty-six day number now circulates in much the way its predecessor did, and it carries its own caveats, which the study states plainly. The sample was small, under a hundred participants, recruited from a university population. Behaviours were self-selected, so the sample of behaviours is not representative. Automaticity was self-reported on a rating scale rather than observed. The plateau estimate is derived from a fitted curve, and for a number of participants no plateau was reached within the study window at all.

Substituting one round number for another repeats the original error in a more defensible costume. What the study supports is that the interval varies substantially by person and by behaviour, and that any single figure will misdescribe most cases.

Benjamin Gardner, who has published with Lally on habit formation in health settings, has made a related point in the applied literature: automaticity is a matter of degree rather than a state that switches on at a particular date, and treating it as a threshold event misrepresents what the measurement instruments record. A rating scale that rises gradually over twelve weeks does not identify a day on which something became a habit, and the modelling used to estimate a plateau is an inference about a curve, not an observation of an event.

That distinction is what both round numbers obscure. The twenty-one day figure and the sixty-six day figure share a grammatical form — a fixed interval after which a state is reached — and the research supports neither the interval nor the state.

Emberline records this filing with the origin identified. There was no experiment behind the twenty-one day figure, and the research that does exist reports a distribution rather than a deadline.